slimflow

Tiny, stdlib-only DAG runner for Python.

Install

pip install slimflow

Usage

from slimflow import Task, Workflow, Runner

extract = Task(name="extract", func=fetch_data)
transform = Task(name="transform", func=clean, depends_on=[extract])
load = Task(name="load", func=write_to_db, depends_on=[transform], max_retries=3)

workflow = Workflow("my-pipeline", [extract, transform, load])
results = Runner(threads=4).run(workflow)

Run artifacts

Pass artifacts_dir= to Runner to persist run_results.json (one per run) at the end of each workflow. Failed or skipped tasks also get a sidecar log file. Below is the schema.

{
  "schema_version": 1,
  "run_id": "8c7a...",
  "workflow_name": "my-pipeline",
  "started_at": "2026-05-23T17:04:11.000000Z",
  "ended_at":   "2026-05-23T17:09:32.000000Z",
  "duration_s": 321.4,
  "python_version": "3.12.4",
  "slimflow_version": "0.1.0",
  "caller_path": "main.py",
  "tasks": [
    {
      "name": "extract",
      "status": "success",
      "attempts": 1,
      "started_at": "2026-05-23T17:04:11.000000Z",
      "ended_at":   "2026-05-23T17:07:55.000000Z",
      "duration_s": 224.0,
      "thread_name": "ThreadPoolExecutor-0_2",
      "depends_on": [],
      "func_module": "my_pipeline.tasks",
      "func_qualname": "fetch_data",
      "func_source_path": "my_pipeline/tasks.py",
      "func_source_line": 12,
      "error": null,
      "traceback": null,
      "log_path": null
    }
  ]
}

Func metadata caveats

func_module / func_qualname / func_source_path are best-effort. Plain functions, functools.partial, and @functools.wraps-decorated functions all resolve cleanly. Lambdas show as "<lambda>". Callable objects fall back to their class name. None of this affects execution — only the BI-grouping tags.

Issues

github.com/whimsical/slimflow/issues